The most human technology we've ever built is being treated like a tool for techies

We were told AI would be for everyone. Yet, look at who actually uses it and it becomes clear that hasn't happened yet. Half of US workers, according to Gallup's latest figures, say they never use AI in their role. Only 13% of them use AI daily - and only basic use at that. While reports frame overall AI use as increasing compared to past quarters, we’re still far from the value of AI being evenly distributed. The real gains from AI are, as I see it every day, landing with a narrow group - technical teams, senior people who like and are used to experimenting, natural early adopters of new tools, and those already flexible and open to innovating in the way they work. Everyone else has an AI tool login they barely open. In some organisations that divide is closing but it’s not doing so fast enough.

That bothers me more than almost anything else in this industry, because AI is the one technology that didn't have to go this way.

It's a genuinely incredible power, and we keep treating it as something for techies, not  the most human technology we've built to date. You don't need clever, fancy language to use it. Anyone can just talk to it. That makes it far closer to electricity than code, and the people that should most get their hands on it are not (primarily) those influenced previous waves of technology. 

There is a recurring frustration. A company buys five hundred ChatGPT licences and calls it their AI strategy. It’s far from strategic. It's five hundred separate experiments, with no shared memory and no connection to each other. The gains show up in isolated pockets, the knowledge and context leaves when the person does, and no gains are really observed by anyone individually or collectively. 

Here's what I actually believe. When a company is going to run on a piece of software, three things should be non-negotiable. You shouldn't be locked to any single model. You should be able to get under the hood and change how it works, which means it has to be open. And because your whole company will end up depending on it, you should be able to look inside and see exactly what it's doing. That belief is why, on June 24th, Mindstone gave a big part of what we've built - Rebel, a system that connects to a company's existing tools and gives every employee their own AI agent, built on a shared organisational memory - away for free, so other people can take it, inspect it and improve on it.  The benefit of AI should accrue to everyone, not just whoever can afford the biggest vendor contract.

Something more difficult I’ve also been reflecting on is that the value of a lot of intellectual work is heading towards zero. When people ask me what my kids should learn, my honest answer is that I assume they may never work the way I have. Long term, I actually believe this is going to be amazing for everyone but short term, while the change is happening, who captures the gains matters enormously, because the gap between people who use this well and people who don't is already something like tenfold. Both of those things can be true at once.

The good news is what happens when you stop gatekeeping it. At one company we work with, AI use worked primarily from the bottom up. One person showed a colleague what their AI could do, the colleague wanted one too, and it spread through the building from there. They ended up calling it ‘the potatoes effect’. Marketing managers and finance analysts were building their own automations in the first week, and none of them were engineers. This, in combination with a real mandate for AI use by leadership, is a very powerful mechanism.

That's the whole point - given the chance, and the right set up and amount of trust, people figure AI out fast.

So if you run a company, my advice is actually a lot simpler than the debate about various AI models and how they compare to each other. Don't hand people a Claude login and sit there waiting to see the benefits. Give them the time, the room, the nudge and the trust, to actually try things. 

Get that right and productivity takes care of itself. Get it wrong and we will have built the most powerful tool in history and handed it to the smallest number of people. That is the choice almost every company will need to confront sooner rather than later. 

The divide is already opening up - on one side, are the organisations where AI is used alone and on the other, teams running on shared memory and systems that learn what to do, when, and with which model. There is still time to capture the gains.